编辑：这并不总是有效：
>>> a,b,c = np.unique(data, return_index=True, return_inverse=True)>>> c # almost!!!array([1, 1, 1, 0, 0, 0, 0, 2, 2, 3, 3, 3])>>> np.argsort(b)[c]array([0, 0, 0, 1, 1, 1, 1, 2, 2, 3, 3, 3], dtype=int64)
但这确实有效：
def replace_groups(data): a,b,c, = np.unique(data, True, True) _, ret = np.unique(b[c], False, True) return ret
并且比字典替换方法更快，对于较大的数据集，大约为33％：
def replace_groups_dict(data): _, ind = np.unique(data, return_index=True) unqs = data[np.sort(ind)] data_id = dict(zip(unqs, np.arange(data.size))) num = np.array([data_id[datum] for datum in data]) return numIn [7]: %timeit replace_groups_dict(lines100)10000 loops, best of 3: 68.8 us per loopIn [8]: %timeit replace_groups_dict(lines200)10000 loops, best of 3: 106 us per loopIn [9]: %timeit replace_groups_dict(lines)10 loops, best of 3: 32.1 ms per loopIn [10]: %timeit replace_groups(lines100)10000 loops, best of 3: 67.1 us per loopIn [11]: %timeit replace_groups(lines200)10000 loops, best of 3: 78.4 us per loopIn [12]: %timeit replace_groups(lines)10 loops, best of 3: 23.1 ms per loop
